poster - Stanford University

Linear Analysis for Determining and Visualizing Critical Thermal
Boundaries of Power Systems
Robert Entriken
1
1
2
Walter Murray
Tomas Tinoco De Rubira
Electric Power Research Institute
2
2
Stanford University
Introduction
Recommended Power Adjustments
Current security assessment practices typically
require computationally expensive system simulations
rely significantly on operator experience
Define measure of system security
X
T
M (y) :=
log(qi y − ti) (N are nearest boundaries)
i∈N
∇M (y) gives power adjustment direction for improving security
may not be practical, can use sparse approximations
However, systems are becoming more stressed and unpredictable, hence
require faster and more frequent security analyses
require considering and monitoring more potentially critical quantities
This work explores
fast techniques based on linear system model for identifying critical
thermal (and voltage) boundaries
visualization techniques for enhancing operator awareness of system
security
1
Model Accuracy
Model accuracy near boundaries tested on two 2.5k-bus systems
2
Modeling Approach
The system model considered is given by
Ax = b + Y y,
Cx ≥ d,
y
min
≤y≤y
max
x is system state, y is vector of generator and load powers
equalities are linearized power flow equations
inequalities are voltage and linearized thermal bounds (based on
current magnitudes)
Updating model with information obtained from linearizing system at new
locations helps reduce errors
Visualization
Can be expressed in terms of y only
Qy ≥ t,
y
min
≤y≤y
max
Strategy
determine a suitable plane P
reconstruct nearest boundaries on
that plane
1
allows expressing boundary distances in terms of power changes
Q is dense, so not constructed!
2
Nearest Boundaries
Identifying nearest boundaries requires comparing distances from current
point y0 (generator and load powers)
requires knowing kqk2 for each row q of Q
T
2
exploit VAR(q z) = kqk2, where z are i.i.d. N (0, 1)
estimate kqk2 through sampling and matrix-vector products with Q
What plane should be used?
One that tries to preserve boundary distances, giving more emphasis to nearest
ones, e.g., “security plane”
P := span{u1, u2},
u1 and u2 are left singular vectors of
Ms := [w1q1 · · · wmqm] ,
qT y = t
wi are weights inversely proportional to
boundary distance.
q
T
q y 0− t
‖q‖2
y0
Experiment: identify 500 nearest boundaries of 45k-bus system (117k
thermal boundaries, 90k voltage boundaries)
algorithm time (s) samples % top 500 % top 50
sampling 1.28 ± 0.09 53 ± 4 95 ± 2 100 ± 0
full Q
4836.34
100
100
Conclusions
Proposed techniques are efficient and effective for identifying and visualizing critical thermal boundaries
Linear system model is accurate for many but not all of the critical thermal boundaries
ISGT Conference
February 18, 2015
Washington, DC